System designers for applications ranging from consumer, automotive, industrial, medical, communications, the Internet of Things (IoT), and the enterprise must account for a multitude of clock timing requirements and performance characteristics, particularly where legacy standards support is required. These include accuracy, precision, stability, system noise, electromagnetic interference (EMI), power consumption, output type (differential or single), and various spread spectrum profiles. The challenge for designers is to meet the various requirements in a small form factor with low power consumption.
At the same time, they must also keep cost and delivery times to a minimum, which is difficult for custom configurations where designers still need to order in production quantities and may face lead times of three to five weeks, or possibly longer. These delays slow both prototyping and development, as well as the final product production schedule.
To meet the need for a more flexible high-performance timing solution, designers can use programmable microelectromechanical systems (MEMS) oscillators instead of classic crystal oscillators. These meet or exceed quality and performance requirements but come in standard structures that can be tuned to meet custom requirements.
This article briefly introduces programmable MEMS oscillators and describes their main elements. It then looks at example devices from SiTime and shows how they can be selected and used to meet the timing requirements for a wide range of applications, while reducing lead times and lowering overall cost.
Why use programmable MEMS oscillators?
Until MEMS oscillators emerged in the 2000s, quartz crystal resonators had dominated circuit timing. However, supported by rapid innovation and the use of silicon processes, MEMS oscillators became a preferred solution where the design requirements emphasized quality, reliability, and robustness. While quartz oscillators remain a good low-cost option for many applications, compared to highly integrated and programmable MEMS devices they can be somewhat more complex to design in. For instance, designers working with quartz oscillators need to choose the correct resonator and load capacitor to avoid issues such as cold start failure and mismatched crystals, while also working to minimize EMI.
The plug-and-play usability of programmable MEMS devices avoids or greatly mitigates these complications. Also, their simple, standardized manufacturing process, combined with their small size, provides inherent performance, reliability, and resilience characteristics. For example, the use of high-volume silicon-based MEMS manufacturing processes minimizes opportunities for contamination, resulting in fewer defective parts per million (DPPM). This lowers cost, but just as important for designers, the process enhances quality and reliability, raising the mean time between failure (MTBF). This applies across environmental temperature extremes, from -55˚C to +125˚C.
With regard to size, MEMS oscillators’ small mass—a standard 32 kilohertz (kHz) MEMS oscillator can be delivered in a chip scale package (CSP) the size of a pin head—means they are extremely robust in the face of shock and vibration. Also, programmable MEMS oscillators have no exposed pc board connections between the resonator and oscillator circuit, and because the oscillator circuits are optimized for electrically noisy conditions, they are much less sensitive to EMI. Their structure and design are such that they are also less sensitive to board noise.
Elements of a programmable MEMS oscillator
The programmable MEMS device comprises a MEMS resonator packaged with a CMOS IC. This CMOS IC contains the analog oscillator control and drive circuitry to generate the required clock (CLK) output (Figure 1). The circuitry typically includes a fractional-N phase lock loop (PLL) and associated dividers, drivers, voltage regulators, and temperature compensation, as well as circuitry to drive the MEMS resonator through electrostatic excitation. The one-time-programmable (OTP) memory shown in Figure 1 is used to store the programmed parameters.
Figure 1: The programmability of MEMS oscillators derives from the use of configurable analog oscillator circuity in a CMOS IC packaged with a MEMS resonator, shown at left (three different types, chosen based on the application). (Image source: SiTime)
Unlike quartz crystal oscillators, where different parts are chosen or manufactured based on the required CLK, programmable MEMS oscillators are manufactured in batches of blanks that are field programmable for the required output frequencies. Along with frequency, other programmable parameters include supply voltage, frequency stability, and rise/fall times, among others (Figure 2).
Figure 2: The wide range of programmable MEMS timing options provides designers with the flexibility to efficiently and cost-effectively meet the needs of multiple generations of systems across a range of applications. (Image source: SiTime)
This parametric fine-tuning allows a designer to program the output frequency to exactly match downstream ICs, such as microcontrollers, microprocessors, or a system-on-chip (SoC). This flexibility, which also eliminates the need for external buffers, frequency dividers, or frequency translation PLLs, greatly reduces complexity and development time.
While programmable MEMS oscillators greatly lessen a designer’s burden, that burden doesn’t disappear. Instead, it moves upstream to the device provider, which designers depend on to have the MEMS, programmable analog, and systems expertise to ensure a reliable and stable solution that is easily programmed.
Programmable MEMS solutions
Though flexible, there is no “one size fits all” option to cover all possible applications across all frequencies. Still, programmable MEMS oscillator processes and technology have been mastered to the point that they can get very close. For example, the SiT3521 (Figure 3) and SiT3522 oscillators from SiTime’s Elite Platform are capable of in-system programmability (ISP) using their I2C/SPI interface over the range of 1 megahertz (MHz) to 340 MHz, and 340 MHz to 725 MHz, respectively, in 1 hertz (Hz) increments.
Figure 3: The SiT3521 (pictured) has a digital I2C/SPI interface (bottom right) and is programmable from 1 MHz to 340 MHz. Its sister device, the SiT3522, is programmable from 340 MHz to 725 MHz. (Image source: Digi-Key Electronics)
As digitally controlled oscillators (DCOs), the devices don’t need a digital-to-analog converter (DAC) to drive the control input, and they aren’t subject to analog noise coupling.
Also, because frequency pulling is achieved through a fractional feedback divider of the PLL, there is no pull nonlinearity. The use of a fractional feedback divider also means the pullability isn’t constrained, as it might be with a voltage-controlled quartz crystal oscillator. This allows the devices to have 16 frequency pull range options between 6.25 parts per million (ppm) to 3200 ppm. Both devices have ultra-low phase jitter of ~0.2 picoseconds (ps) and specified programmable pull ranges from ±25 ppm up to ±3200 ppm. Their frequency pull resolution is as low as 5 parts per trillion (ppt), and they support three signaling types: LVPECL, LVDS, and HCSL.
Their flexibility makes the devices suited to applications such as networking, server storage, broadcasting, telecom, and test and measurement. Here, the need for backward compatibility with legacy standards, such as for digital video transmission or Ethernet, requires the ability to accommodate multiple frequencies, as well as various jitter and phase noise requirements.
Using the SiT3521 and SiT3522 programmable MEMS oscillators
In operation, the SiT3521 and SiT3522 have two modes: “any frequency” and DCO. In any frequency mode, designers can reprogram the device to any of its supported frequencies. To accomplish this, they need to first calculate the post-divider, feedback, and mDriver values, and then write them to the device (Figure 4).
Figure 4: Referring to a high-level block diagram of the I2C/SPI oscillator, programming of both the SiT3521 and SiT3522 starts with the calculation of the post-divider, feedback divider, and mDriver values, with the one user input value for these calculations being the target output frequency. (Image source: SiTime)
The only input value from the designer that is required for these calculations is the required output frequency. The other input values are the divider’s allowed ranges. Note that when a new value is programmed, the output is disabled for a short time, so the designer needs to account for that.
For digital control, the process is easier. The device powers up to its nominal operating frequency and pull range, per the device’s ordering code. From that point, both the pull range and output frequency can be set by writing to their respective control registers (upper left, Figure 4). There are, however, some nuances to consider. For example, the maximum output frequency change is constrained by the pull range limits. The pull range is specified as half the peak-to-peak deviation, so a deviation of 200 ppm peak-to-peak is specified as a pull range of ±100 ppm.
After choosing the required pull range from the list of 16 options (between ±6.25 ppm to ±3200 ppm, mentioned earlier), the pull range is loaded to the respective control register (Reg2[3:0], Figure 4). The pull range affects the frequency precision, per Table 1.
|Reg2[3:0]||Programmed Pull Range||Frequency Precision|
Table 1: Designers can select their choice of 16 possible SiT3521 and SiT3522 pull ranges and load it into the control register. The choice of pull range affects the frequency precision. (Image source: SiTime)
To change the output frequency, the designer writes two control words: first the least significant word (LSW) to Reg0[15:0], followed by the most significant word (MSW) to Reg0[15:0]. After the MSW is written, the device changes its feedback divider value to accommodate the new frequency. This is done during the Tdelay timeframe (Figure 5).
Figure 5: In DCO mode, the output frequency change is initiated after the MSW is written, and is finished after the device changes its feedback value (during Tdelay) and settles (Tsettle) to 1% of its new value (F1). (Image source: SiTime)
After the divider value is set, the output settles to within 1% of the final frequency value. Unlike the “any frequency” mode, the output is not disabled during frequency changes. However, if the software output enable (OE) control function is enabled, the designer can choose to disable the output manually during the frequency-change period.
To get comfortable with the devices and ensure they meet application requirements, designers can experiment with them using the SiT6712EB evaluation board. It supports both the SiT3521 and SiT3522 with differential signaling outputs in the 10-pin QFN package and allows evaluation of all aspects of the devices, including signal integrity, phase noise, phase jitter, and ease of reprogramming. It supports LVPECL, LVDS, and HCSL output signal types and includes probing points for output frequency measurements.
It’s important to point out here that these are differential oscillators with sub-nanosecond rise/fall times. To ensure accurate measurements, it’s important to use measurement best practices, along with a high-quality active probe (Figure 6).
Figure 6: When using the SiT6712EB evaluation board, it’s important to employ high-speed measurement best practices, including the use of a high-quality active probe and suitable high-speed differential probe heads. (Image source: SiTime)
For best results, an active probe with a bandwidth of >4 gigahertz (GHz) and a load capacitance of <1 picofarad (pF) should be used, with matching high-speed differential probe heads. The accompanying oscilloscope should have a bandwidth of 4 GHz or higher, with 50 ohm (Ω) inputs.
Application-oriented, off-the-shelf programmable oscillators
There are of course many series of programmable MEMS oscillators, and while some are suited for networking, broadcast, and communications, others may have characteristics that make them suitable for automotive, such as AEC-Q100 qualification, or industrial, with an emphasis on features such as a high operating temperature range. For example, the SiT1602BI-33-33S-33.333330 has an operating temperature of -40˚C to +85˚C; 33.333330 denotes its nominal frequency in megahertz.
There are package and voltage options, too, that are suited to a particular application. For example, the SiT1532 is a low-voltage CMOS (LVCMOS) 1.2-volt oscillator in a UFBGA package with a footprint of 1.54 millimeters (mm) x 0.84 mm, and a height of 0.60 mm (Figure 7). Targeting mobile and IoT applications, it has a nominal frequency of 32.768 kHz.
Figure 7: The SiT1532 is an LVCMOS programmable MEMS oscillator in a UFBGA package for IoT and mobile applications. (Image source: SiTime)
For automotive, the SiT8924AE 24 MHz oscillator combines a very high operating temperature range (-55˚C to ~125˚C) and a small, no-lead surface mount device (SMD) package with a footprint of 2.50 mm x 2.00 mm, and a height of 0.80 mm.
While these programmable MEMS devices, comprising dozens of series, are readily available off the shelf in their nominal frequencies, all have the same original form: blanks. They are essentially “field programmable” oscillators that started as blanks that were then pre-programmed at the factory for commonly required frequencies and stocked by Digi-Key.
Rapid shipping of custom oscillators
Having a wide variety of oscillators available helps get commonly required timing circuits to market quickly, but not every designer wants to deal with programming the oscillator, despite it being fairly straightforward, and in certain cases, they may also need custom configurations. Historically, the latter would have meant a three to five-week lead time for the custom configuration to ship from the factory. Digi-Key tackled this problem by installing an automated programming machine—dedicated to SiTime parts—in its own warehouse (Figure 8).
Figure 8: Digi-Key’s automated programming machine, dedicated to SiTime oscillators, is shown placing a blank oscillator in its programming socket. (Image source: Digi-Key Electronics)
The machine currently has eight sockets and can program up to 1500 units per hour, reducing the lead time for custom configurations to 24 to 48 hours, with no minimum quantities.
To take advantage of this capability, designers start at the SiTime Programmable Oscillators section on Digi-Key’s TechForum. Once a request is submitted, an e-mail will be immediately sent to one of Digi-Key’s Engineering Technicians. They will verify the new part number and get it added to the Digi-Key website. While the website will guide designers through the ordering process, familiarity with SiTime nomenclature for its oscillator configurations can be helpful (Figure 9).
Figure 9: Shown is the configuration nomenclature typically used for SiTime programmable MEMS oscillators, in this case for the SiT2001 base model. (Image source: SiTime)
Designers of systems for a range of applications need flexible circuit timing solutions to meet current—as well as legacy and future—system specifications and requirements. Instead of multiple crystal or MEMS oscillators and the associated circuits and design complexities, designers can save space, time, and cost by opting for programmable MEMS devices that can already meet many of their requirements.
If custom designs are required, designers don’t have to wait three to five weeks for production shipment from the factory. Using a programming machine dedicated to SiTime devices, Digi-Key can start shipping custom configurations in 24 to 48 hours.